Time scaling algorithm of speech signal to assist learning of a foreign language

For basic level students of foreign languages, normal speed speaking is very difficult to understand perfectly, because of lack of training in understanding of oral language. However when the speed of speaking slows down, in most cases understanding increases. This fact suggests that to improve learning of the foreign language, it is necessary that students can adjust the speed of speaking according to their own understanding level. This paper presents a comparison of two time scaling algorithms when they are used to assist learning of a foreign language. Both algorithms consist of a pitch detection stage and time scaling stage. The pitch detection of both algorithms is based on autocorrelation method of the speech signals proposed by Rabiner et. al. (1976). The time scaling in the first method consists in duplicating the pitch periods of voiced segments while keeping unchanged unvoiced ones. The second method is based on the short time Fourier transform. Experimental results, MOS (mean opinion scoring) are given using Spanish, French, German, Russian, Japanese and Italian which show desirable features of both time scaling algorithms when they are used to assist the students to learn foreign languages. The performance of both algorithms when the pitch detection stage has some noise is also shown